SDKs & Integration
There are two ways to connect an agent to Clausgap. Pick whichever matches how your agent is actually built — both end up enforcing the same policies.
Not using a recognized framework?
Direct HTTP integration →
Route your agent's calls through the proxy and sign the X-Clausgap-Context header yourself, or use the core Python/TypeScript SDK, which handles signing, escrow polling, and rate-limit errors for you. Works with any tool your agent can reach over HTTP.
Already using an agent framework?
LangChain & LangGraph →
Callback handler, @guarded_tool decorator, or @guarded_node for LangGraph.
CrewAI →
ClausgapHooks intercepts every tool call across a whole crew.
OpenAI Agents SDK →
ClausgapGuard wraps @function_tool tools before they're handed to an Agent.
Anthropic Agent SDK →
ClausgapRunHooks plugs into the Claude Agent SDK's run_hooks.
AWS Bedrock Agents →
Lambda action-group handler, or a standalone check for RETURN_CONTROL mode.
Vercel AI SDK →
Guard a single tool's execute, or every tool call via a model middleware.
Building something that receives events from Clausgap instead — an escrow hold created, a policy approved — rather than an agent that calls out? See webhook verification →